Weird behavior for atomic load

[zzjas]

Describe the bug

i32.atomic.load won't give a runtime error for out-of-bounds access.

Steps to reproduce

(module
    (memory 1)
    (func (export "_start")
        i32.const 0xffff_fff0
        i32.atomic.load offset=16
        drop
    )
)

Run with wasmer --enable-all atomic.wat

Expected behavior

We think this should be a runtime OOB error but wasmer will return normally.

If the 0xffff_fff0 is changed to 0xfff0, wasmer will then give OOB error. Or if the i32.atomic.load is changed to i32.load, wasmer will also give the error.

Please excuse us if this behavior is intended. Also, the test is collected from bytecodealliance/wasmtime#3132 in case the discussion there might be useful.

Thanks.

[zzjas]

Sorry forgot to add this. It was tested using
wasmer version: 2.3.0
OS: macOS 12.5.1 (M1 Chip)

[ptitSeb]

What compiler are you using? Cranelift (the default one) or Singlepass?

[zzjas]

Cranelift

[ptitSeb]

Ok, thanks. I'll have a look at it. It seems there is the same "warp around" issue as with wasmtime.

[ptitSeb]

Merged